An intelligent cockpit control method for agricultural machinery control

Abstract

The present invention discloses an intelligent cockpit control method for agricultural machinery control, the method comprising data collection, data preprocessing, operating environment analysis, cockpit comprehensive adjustment and intelligent cockpit control. The present invention relates to the field of intelligent agricultural machinery control technology, specifically to an intelligent cockpit control method for agricultural machinery control, the present invention adopts an intelligent cockpit control method that combines operating environment analysis and cockpit comprehensive adjustment, by combining operating environment analysis and cockpit adjustment decision optimization, the intelligence level of cockpit control is improved, and the standardization and universality of the entire cockpit control decision are also improved; an exogenous long short-term memory neural network combined with expert model hybrid graph convolution is used to perform operating environment analysis; and an integrated optimized multi-agent reinforcement learning method is used to perform cockpit comprehensive adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent agricultural machinery control, and in particular to an intelligent cockpit control method for agricultural machinery control. Background Art

[0002] The intelligent cockpit control method for agricultural machinery control is an intelligent control technology that integrates advanced sensors, artificial intelligence algorithms and automatic control systems to monitor the operator's physiological state, environmental changes and operating parameters in real time and intelligently adjust various comfort and safety factors in the cockpit. These adjustment factors include seat position, temperature and humidity, etc., and at the same time, combined with the operator's fatigue state and work progress, provide the operator with the best working environment, improve work efficiency, reduce fatigue, and reduce the incidence of human errors. The core role of this method is to optimize the operator's work experience and health through automated adjustment and personalized adaptation, thereby improving the overall efficiency and safety of agricultural machinery operations.

[0003] However, among the existing intelligent agricultural machinery control methods, there are existing intelligent methods that focus more on the macro-adjustment of the agricultural machinery itself or the optimization of the operation mode, while the technical ideas of intelligent agricultural machinery control not only focus on specific agricultural machinery operations or machine adjustments, but also on the control and adjustment of the agricultural machinery cockpit, which also plays an important role in improving the operation efficiency and the operator's operation process. The existing technology for the comprehensive control of the agricultural machinery cockpit often relies on the personal experience of the operator, or direct adjustment based on fixed rules, which leads to the lack of technical standardization and universality of the existing methods, and the technical problem that the intelligence needs to be improved; in the existing working environment analysis methods, there are some problems that the agricultural machinery working environment time involves extremely long working hours. The environment is complex and requires a large amount of environmental monitoring data. The existing analysis methods for the operating environment basically use a single model for simple time series prediction. However, this method sacrifices the evaluation of complex situations and cannot meet the subsequent refined needs for cockpit adjustment. In the existing cockpit comprehensive adjustment methods, there are control decision methods themselves. Recent technical solutions tend to use reinforcement learning methods for intelligent decision optimization. However, the comprehensive adjustment of the cockpit involves both the mechanical control of the cockpit itself and the adjustment of the physical environment inside the cockpit. Therefore, the traditional reinforcement learning methods have technical problems that need to be improved in terms of accuracy and learning standardization. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent cockpit control method for agricultural machinery control. In view of the fact that the existing intelligent agricultural machinery control methods are mostly focused on the macro-adjustment of the agricultural machinery itself or the optimization of the operation mode, the technical idea of ​​intelligent agricultural machinery control not only focuses on the specific agricultural machinery operation or machine adjustment, but also plays an important role in improving the operation efficiency and the operator's operation process for the control and adjustment of the agricultural machinery cockpit. The prior art for the comprehensive control of the agricultural machinery cockpit often relies on the personal experience of the operator, or is a direct adjustment based on fixed rules, which leads to the technical standardization and universality of the existing methods. At the same time, the intelligence needs to be improved. The present invention creates a technical solution. The intelligent cockpit control method that integrates working environment analysis and comprehensive cockpit adjustment is adopted. By combining working environment analysis and cockpit adjustment decision optimization, the intelligence level of cockpit control is improved, and the standardization and universality of the entire cockpit control decision are also improved. In the existing working environment analysis methods, the working environment of agricultural machinery involves extremely complex environmental change factors, and a large amount of environmental monitoring data needs to be detected. The existing analysis methods for the working environment basically use a single model for simple time series prediction, but this method sacrifices the evaluation of complex situations and cannot meet the subsequent technical problems of refined cockpit adjustment. This scheme creatively adopts an exogenous long-range method combined with an expert model hybrid graph convolution Short-term memory neural network is used to analyze the working environment. By constructing a mixture of expert model and graph convolution model, and introducing the operator's adjustment data as exogenous data at this stage, the connection between the results of working environment analysis and prediction and cockpit adjustment is enhanced. At the same time, the design of expert model and graph convolution model enhances the correlation of time and space features, and improves the accuracy and reliability of working environment analysis as a whole. In the existing cockpit comprehensive adjustment method, there is a problem with the control decision method itself. Recent technical solutions tend to optimize intelligent decision-making through reinforcement learning methods. However, cockpit comprehensive adjustment involves both mechanical control of the cockpit itself and adjustment of the physical environment inside the cockpit. Therefore, traditional reinforcement learning methods are limited in accuracy and There are technical issues that need to be improved in terms of learning standardization. This solution creatively adopts an integrated optimized multi-agent reinforcement learning method to perform comprehensive cockpit adjustment. By integrating the optimization ideas of two optimization algorithms, proximal policy optimization and deep deterministic policy gradient optimization, deep deterministic policy gradient optimization is used to improve the accuracy of continuous adjustment action decisions. Proximal policy optimization is used to ensure that the adjustment strategy will not change drastically, thereby improving the stability of the entire learning process. At the same time, the multi-agent design also optimizes the adaptability of reinforcement learning applied to comprehensive cockpit adjustment. By adjusting multiple levels of mechanical and physical environment, the overall intelligence of the comprehensive cockpit adjustment method is improved, and strong support and assistance is provided for intelligent agricultural machinery control.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent cockpit control method for agricultural machinery control, the method comprising the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: operating environment analysis;

[0009] Step S4: comprehensive cockpit adjustment;

[0010] Step S5: Intelligent cockpit control.

[0011] Furthermore, in step S1, the data collection is used to collect the raw data required for the intelligent cockpit control, specifically, by collecting the raw data set for the intelligent cockpit control from the agricultural machinery operation process data through sensor arrangement and feedback collection from the agricultural machinery operator;

[0012] The smart cockpit controls the original data set, specifically including environmental data, mechanical data, operator physiological data and operator input data.

[0013] Further, in step S2, the data preprocessing is used to perform basic quality optimization and processing on the original data, specifically, optimizing the original data set of the intelligent cockpit control to obtain an optimized cockpit control data set, which specifically includes the following steps:

[0014] Step S21: data cleaning, which is used to eliminate erroneous data, specifically, manually checking the numerical data in the original data set of the smart cockpit control and filling the missing data or the erroneous data after elimination by using the mean, and manually checking the categorical data in the original data set of the smart cockpit control and filling the missing data or the erroneous data after elimination by using the mode to obtain cleaned and optimized data;

[0015] Step S22: outlier processing, specifically, using a statistical method to detect outliers in the cleaned and optimized data, and filling the data after the outliers are removed by mean filling to obtain outlier optimized data;

[0016] Step S23: data standardization, specifically, using type data coding to perform a standardization operation on the classified data in the outlier optimization data to obtain standardized data;

[0017] Step S24: data normalization, specifically, using maximum and minimum normalization to map the numerical data in the outlier optimization data to the interval [0, 1] to obtain normalized data;

[0018] Step S25: feature initialization, specifically, performing initial feature extraction through timestamp processing and statistical feature extraction to obtain initial feature data, wherein the initial feature data specifically includes mean feature, maximum feature, change rate feature and vibration data frequency domain feature;

[0019] Step S26: data integration and fusion, specifically, performing data integration and fusion through the data cleaning, the outlier processing, the data standardization, the data normalization and the feature initialization, and performing a timestamp alignment operation to obtain an optimized cockpit control data set;

[0020] The optimized cockpit control data set specifically includes optimized environmental data, optimized mechanical data, optimized operator physiological data and optimized operator input data.

[0021] Further, in step S3, the working environment analysis is used to perform basic analysis on the working environment of the agricultural machinery, specifically, based on the optimized cockpit control data set, an exogenous long short-term memory neural network combined with a hybrid graph convolution of an expert model is used to perform working environment analysis to obtain working environment analysis reference data;

[0022] The exogenous long short-term memory neural network combined with the expert model hybrid graph convolution specifically includes an expert model hybrid graph convolution subnet and an exogenous long short-term memory subnet;

[0023] The expert model hybrid graph convolution subnet is used to extract the operation environment factor graph information from the optimized environment data and the optimized mechanical data in the optimized cockpit control data set, and capture the correlation characteristics between the environmental factors through the graph convolution operation;

[0024] The exogenous long short-term memory subnet is used to combine the optimized operator physiological data and the optimized operator input data in the optimized cockpit control data set as external data, and predict the future dynamic change trend of the environment by combining the correlation characteristics between environmental factors;

[0025] The step of using an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform work environment analysis and obtain work environment analysis reference data includes:

[0026] Step S31: constructing a graph structure, specifically, constructing a working environment graph feature matrix and an environmental factor adjacency matrix according to the optimized environment data and optimized mechanical data in the optimized cockpit control data set, to obtain a working environment graph data set;

[0027] Step S32: constructing a graph convolutional subnetwork, specifically, constructing three parallel standard graph convolutional neural networks based on the work environment graph dataset, and obtaining work environment related feature data through node aggregation and node relationship feature extraction;

[0028] Step S33: expert model construction, specifically, constructing an expert model to adjust the weight calculation data of the working environment related feature data, and by constructing three expert models, respectively optimizing the feature extraction effects of three parallel standard graph convolutional neural networks, and obtaining the expert model optimized environment related feature data;

[0029] Step S34: constructing an expert model hybrid graph convolutional subnet, specifically, constructing an expert model hybrid graph convolutional subnet through the graph structure construction, the graph convolutional subnet construction and the expert model construction, and performing a graph convolution operation by applying the expert model hybrid graph convolutional subnet to obtain environmental spatiotemporal correlation feature data;

[0030] Step S35: constructing an exogenous long-short-term memory subnet, specifically, taking the environment spatiotemporal correlation feature data as internal data, and introducing the optimized environment data and optimized mechanical data in the optimized cockpit control data set as external data, constructing an exogenous long-short-term memory subnet, and performing temporal feature processing to obtain a comprehensive output of the operating environment;

[0031] Step S36: training the work environment analysis model, specifically by combining the expert model hybrid graph convolution subnet and the exogenous long short-term memory subnet into a joint model, processing the spatial and temporal features of the environmental data and training the work environment analysis model to obtain the work environment analysis model Model WE ;

[0032] Step S37: Work environment analysis, specifically using the work environment analysis model Model WE , performing operating environment analysis based on the optimized cockpit control data set to obtain operating environment analysis reference data;

[0033] The working environment analysis reference data specifically includes working environment state mode prediction results, working environment change trend prediction results and environmental anomaly detection prediction results;

[0034] The working environment state model prediction result specifically includes the state classification of the current working environment and the specific impact weights of environmental factors;

[0035] The prediction result of the change trend of the working environment specifically refers to the change trend of the environmental variables within the next hour;

[0036] The environmental anomaly detection prediction result specifically refers to the prediction result of whether the current operating environment is abnormal or not.

[0037] Further, in step S4, the cockpit comprehensive adjustment is used to perform intelligent cockpit adjustment in combination with the agricultural machinery working environment and the agricultural machinery comprehensive data, specifically, based on the working environment analysis reference data and the optimized cockpit control data set, an integrated optimized multi-agent reinforcement learning method is used to perform cockpit comprehensive adjustment to obtain cockpit comprehensive adjustment reference data, specifically including the following steps:

[0038] Step S41: constructing a multi-agent environment, specifically using a multi-agent design to perform reinforcement learning agent definition, wherein the multi-agent design specifically includes a seat adjustment agent, a temperature adjustment agent, and a humidity adjustment agent;

[0039] Step S42: constructing a reinforcement learning parameter space, specifically constructing a reinforcement learning parameter combination, specifically including state space parameters, action space parameters and reward function parameters;

[0040] The state space parameters specifically include the work environment analysis reference data, operator physiological data and operator input data;

[0041] The action space parameters specifically include seat adjustment action, temperature adjustment action and humidity adjustment action;

[0042] The reward function parameters are specifically designed to calculate the reward value parameters, and the calculation formula is:

[0043] ;

[0044] In the formula, R r is the reward value parameter, is the comfort weight, So is the comfort change value, which is specifically calculated from the optimized operator physiological data, is the operating efficiency weight, Ra is the operating efficiency change value, which is specifically calculated from the operating environment analysis reference data, is the environmental suitability weight, Ev is the environmental suitability parameter, which is specifically calculated from the operating environment analysis reference data, is the adjustment frequency weight, Tr is the adjustment frequency parameter, which is specifically calculated from the optimization operator input data;

[0045] Step S43: Deep deterministic policy gradient optimization is used to optimize the continuous cockpit adjustment control process, specifically to construct a standard deep deterministic policy gradient optimization algorithm, optimize state space parameters and action space parameters, and optimize the learning effect of multi-agents through the experience replay mechanism;

[0046] Step S44: proximal strategy optimization, which is used to stabilize the decision results of the cockpit adjustment control as a whole, specifically, to construct a standard proximal strategy optimization algorithm and optimize the decision stability of multi-agent reinforcement learning by introducing a shear objective function;

[0047] Step S45: reinforcement learning integrated optimization, specifically, performing reinforcement learning integrated optimization through the deep deterministic policy gradient optimization and the proximal policy optimization, and performing multi-agent reinforcement learning comprehensive optimization by constructing an integrated optimization objective function to obtain a multi-agent comprehensive optimization output, and the calculation formula is:

[0048] ;

[0049] In the formula, is the multi-agent comprehensive optimization output, where is the parameter to be optimized, which is used to represent the specific cockpit comprehensive adjustment decision, E r [·] is the expected function, r is the iteration index, min(·) is the minimum function, The whole is used to represent the objective function part of the proximal strategy optimization, where is the strategy ratio, which is used to represent the reward value ratio between the new cockpit adjustment decision and the old cockpit adjustment decision, Advantage function, used to indicate the superiority of the current action, The whole is the objective function part of deep deterministic policy gradient optimization, where: is the balance coefficient, used as the weight for adjusting the integrated optimization, is the Q-value function, which is used to evaluate the current state space parameter s r and action space parameter a r The decision effect, s r is the state space parameter, a r is the action space parameter;

[0050] Step S46: training the cockpit comprehensive adjustment model, specifically, training the cockpit comprehensive adjustment model by constructing the multi-agent environment model, constructing the reinforcement learning parameter space, performing the deep deterministic policy gradient optimization and the reinforcement learning integrated optimization, and obtaining the cockpit comprehensive adjustment model Model SC ;

[0051] Step S47: Comprehensive cockpit adjustment, specifically using the cockpit comprehensive adjustment model Model SC , performing cockpit comprehensive adjustment according to the operating environment analysis reference data and the optimized cockpit control data set to obtain cockpit comprehensive adjustment reference data;

[0052] The cockpit comprehensive adjustment reference data specifically includes cockpit adjustment reference parameters, operator comfort evaluation parameters, work efficiency improvement parameters and environmental comfort index parameters;

[0053] The cabin adjustment reference parameters specifically include seat adjustment parameters, temperature adjustment parameters and humidity adjustment parameters;

[0054] The operator comfort evaluation parameter is specifically used to evaluate the operator's current cockpit comfort by detecting the operator's physiological data;

[0055] The operating efficiency improvement parameter specifically refers to a predicted parameter of operating efficiency improvement after comprehensive cockpit adjustment based on the cockpit adjustment reference parameter;

[0056] The environmental comfort index parameter specifically refers to a comfortable working environment prediction parameter of the current cockpit environment.

[0057] Furthermore, in step S5, the intelligent cockpit control is used to integrate environmental analysis and cockpit adjustment strategy to make comprehensive control decisions. Specifically, intelligent cockpit control is performed through the operating environment analysis and the cockpit comprehensive adjustment, and based on the operating environment analysis reference data and the cockpit comprehensive adjustment reference data, to obtain intelligent cockpit control decision reference data.

[0058] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0059] (1) Among the existing intelligent agricultural machinery control methods, the existing intelligent methods mostly focus on the macro-adjustment of the agricultural machinery itself or the optimization of the operation mode. The technical ideas of intelligent agricultural machinery control not only focus on specific agricultural machinery operations or machine adjustments, but also on the control and adjustment of the agricultural machinery cockpit, which also plays an important role in improving the operation efficiency and the operator's operation process. The existing technology for the comprehensive control of the agricultural machinery cockpit often relies on the personal experience of the operator or direct adjustment based on fixed rules. This leads to the technical standardization and universality of the existing methods, and the technical problem that the intelligence needs to be improved. This scheme creatively adopts an intelligent cockpit control method that integrates the operation environment analysis and the cockpit comprehensive adjustment. By combining the operation environment analysis and the cockpit adjustment decision optimization, the intelligence of the cockpit control is improved, and the standardization and universality of the entire cockpit control decision are also improved;

[0060] (2) In the existing working environment analysis methods, the working environment of agricultural machinery involves extremely complex environmental change factors and requires the detection of a large amount of environmental monitoring data. The existing analysis methods for the working environment basically use a single model for simple time series prediction, but this method sacrifices the evaluation of complex situations and cannot meet the subsequent technical problems of refined cockpit adjustment. This scheme creatively uses an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform working environment analysis. By constructing a hybrid of an expert model and a graph convolution model, and introducing the operator's adjustment data as exogenous data at this stage, the connection between the results of working environment analysis and prediction and cockpit adjustment is enhanced. At the same time, the design of the expert model and the graph convolution model enhances the correlation between time and space features, and overall improves the accuracy and reliability of working environment analysis.

[0061] (3) In the existing cockpit comprehensive adjustment methods, there are problems with the control decision method itself. Recent technical solutions tend to use reinforcement learning methods to optimize intelligent decisions. However, the cockpit comprehensive adjustment involves both the mechanical control of the cockpit itself and the adjustment of the physical environment inside the cockpit. Therefore, the traditional reinforcement learning method has technical problems that need to be improved in terms of accuracy and learning standardization. This solution creatively adopts an integrated optimization multi-agent reinforcement learning method to perform cockpit comprehensive adjustment. By integrating the optimization ideas of two optimization algorithms, proximal policy optimization and deep deterministic policy gradient optimization, deep deterministic policy gradient optimization is used to improve the accuracy of continuous adjustment action decisions. Proximal policy optimization is used to ensure that the adjustment strategy will not change drastically, thereby improving the stability of the entire learning process. At the same time, the multi-agent design also optimizes the adaptability of reinforcement learning applied to cockpit comprehensive adjustment. By adjusting multiple levels of mechanical and physical environment, the overall intelligence of the cockpit comprehensive adjustment method is improved, and it also provides strong support and assistance for intelligent agricultural machinery control. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a flow chart of an intelligent cockpit control method for agricultural machinery control provided by the present invention;

[0063] Figure 2 This is a schematic diagram of the process of data preprocessing in step S2;

[0064] Figure 3 A schematic diagram of the process of operating environment analysis in step S3;

[0065] Figure 4 This is a flow chart of the cockpit comprehensive adjustment process in step S4.

[0066] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0068] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0069] Example 1, see Figure 1 The present invention provides an intelligent cockpit control method for agricultural machinery control, the method comprising the following steps:

[0070] Step S1: data collection;

[0071] Step S2: data preprocessing;

[0072] Step S3: operating environment analysis;

[0073] Step S4: comprehensive cockpit adjustment;

[0074] Step S5: Intelligent cockpit control.

[0075] By performing the above operations, it is found that in the existing intelligent agricultural machinery control methods, the existing intelligent methods focus more on the macro-adjustment of the agricultural machinery itself or the optimization of the operation mode, and the technical idea of ​​intelligent agricultural machinery control not only focuses on specific agricultural machinery operations or machine adjustments, but also the control and adjustment of the agricultural machinery cockpit also plays an important role in improving the operation efficiency and the operator's operation process. The existing technology for the comprehensive control of the agricultural machinery cockpit often relies on the personal experience of the operator, or direct adjustment based on fixed rules, which leads to the lack of technical standardization and universality of the existing methods, and the technical problem that the intelligence needs to be improved. This scheme creatively adopts an intelligent cockpit control method that integrates the operating environment analysis and the comprehensive adjustment of the cockpit. By combining the operating environment analysis and the cockpit adjustment decision optimization, the intelligence level of the cockpit control is improved, and the standardization and universality of the entire cockpit control decision are also improved.

[0076] Example 2, see Figure 1 In step S1, the data collection is used to collect the raw data required for intelligent cockpit control, specifically, by collecting the raw data set of intelligent cockpit control from the agricultural machinery operation process data through sensor arrangement and feedback collection by agricultural machinery operators;

[0077] The smart cockpit control raw data set specifically includes environmental data, mechanical data, operator physiological data and operator input data;

[0078] The sensor arrangement is specifically an arrangement of environmental sensors, mechanical sensors and operator status sensors;

[0079] The agricultural machinery operator feedback collection is specifically to collect agricultural machinery operator feedback by collecting operation records and control feedback records of agricultural machinery operators;

[0080] The environmental data specifically includes soil moisture data, air temperature and humidity data, meteorological data, and terrain slope data of the operation area;

[0081] The mechanical data specifically includes engine temperature data, engine oil pressure data, agricultural machinery speed data, agricultural machinery vibration data and agricultural machinery oil quantity data;

[0082] The operator's physiological data specifically includes heart rate monitoring data, eye movement monitoring data and body sensation monitoring data;

[0083] The operator input data specifically includes seat adjustment records, climate control setting records, and work mode adjustment records;

[0084] The seat adjustment record specifically includes a seat position adjustment record and a tilt angle adjustment record;

[0085] The climate control setting record specifically includes air conditioning temperature, wind speed and seat temperature status adjustment records.

[0086] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to perform basic quality optimization and processing on the original data, specifically, optimizing the original data set of the intelligent cockpit control to obtain an optimized cockpit control data set, which specifically includes the following steps:

[0087] Step S21: data cleaning, which is used to eliminate erroneous data, specifically, manually checking the numerical data in the original data set of the smart cockpit control and filling the missing data or the erroneous data after elimination by using the mean, and manually checking the categorical data in the original data set of the smart cockpit control and filling the missing data or the erroneous data after elimination by using the mode to obtain cleaned and optimized data;

[0088] Step S22: outlier processing, specifically, using a statistical method to detect outliers in the cleaned and optimized data, and filling the data after the outliers are removed by mean filling to obtain outlier optimized data;

[0089] Step S23: data standardization, specifically, using type data coding to perform a standardization operation on the classified data in the outlier optimization data to obtain standardized data;

[0090] Step S24: data normalization, specifically, using maximum and minimum normalization to map the numerical data in the outlier optimization data to the interval [0, 1] to obtain normalized data;

[0091] Step S25: feature initialization, specifically, performing initial feature extraction through timestamp processing and statistical feature extraction to obtain initial feature data, wherein the initial feature data specifically includes mean feature, maximum feature, change rate feature and vibration data frequency domain feature;

[0092] Step S26: data integration and fusion, specifically, performing data integration and fusion through the data cleaning, the outlier processing, the data standardization, the data normalization and the feature initialization, and performing a timestamp alignment operation to obtain an optimized cockpit control data set;

[0093] The optimized cockpit control data set specifically includes optimized environmental data, optimized mechanical data, optimized operator physiological data and optimized operator input data.

[0094] Example 4, see Figure 1 and Figure 3, this embodiment is based on the above embodiment. In step S3, the working environment analysis is used to perform basic analysis on the working environment of the agricultural machinery. Specifically, based on the optimized cockpit control data set, an exogenous long short-term memory neural network combined with a hybrid graph convolution of an expert model is used to perform working environment analysis to obtain working environment analysis reference data;

[0095] The exogenous long short-term memory neural network combined with the expert model hybrid graph convolution specifically includes an expert model hybrid graph convolution subnet and an exogenous long short-term memory subnet;

[0096] The expert model hybrid graph convolution subnet is used to extract the operation environment factor graph information from the optimized environment data and the optimized mechanical data in the optimized cockpit control data set, and capture the correlation characteristics between the environmental factors through the graph convolution operation;

[0097] The exogenous long short-term memory subnet is used to combine the optimized operator physiological data and the optimized operator input data in the optimized cockpit control data set as external data, and predict the future dynamic change trend of the environment by combining the correlation characteristics between environmental factors;

[0098] The step of using an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform work environment analysis and obtain work environment analysis reference data includes:

[0099] Step S31: constructing a graph structure, specifically, constructing a working environment graph feature matrix and an environmental factor adjacency matrix according to the optimized environment data and optimized mechanical data in the optimized cockpit control data set, to obtain a working environment graph data set;

[0100] Step S32: Graph convolution subnetwork construction, specifically, constructing three parallel standard graph convolution neural networks based on the work environment graph dataset, and obtaining work environment related feature data through node aggregation and node relationship feature extraction. The calculation formula is:

[0101] ;

[0102] In the formula, H L+1 is the graph convolution output of the L+1th layer, which is used to represent the associated feature data of the work environment. ReLU(·) is a nonlinear activation function, A is the adjacency matrix of environmental factors, and H L is the graph convolution output of layer L, W L is the graph convolution weight of the Lth layer;

[0103] Step S33: expert model construction, specifically, constructing an expert model to adjust the weight calculation data of the working environment related feature data, and by constructing three expert models, respectively optimizing the feature extraction effects of three parallel standard graph convolutional neural networks, and obtaining the expert model optimized environment related feature data, the calculation formula is:

[0104] ;

[0105] In the formula, H i is the feature output of the i-th parallel standard graph convolutional neural network after expert model optimization, which is used to represent the expert model optimization environment associated feature data, i is the parallel standard graph convolutional neural network index, J is the total number of expert models, and the specific value is 3. When j is 1, it represents the first expert model, and the multilayer perceptron model is used as the basic model. When j is 2, it represents the second expert model, and the standard convolutional neural network is used as the basic model. When j is 3, it represents the third expert model, and the standard self-attention transformer model is used as the basic model. ij is the expert model weight of the jth expert model corresponding to the i-th parallel standard graph convolutional neural network, e j is the optimal expert model, obtained through cross-validation. is the graph convolution output of the i-th parallel graph convolutional neural network;

[0106] Step S34: constructing an expert model hybrid graph convolutional subnet, specifically, constructing an expert model hybrid graph convolutional subnet through the graph structure construction, the graph convolutional subnet construction and the expert model construction, and performing a graph convolution operation by applying the expert model hybrid graph convolutional subnet to obtain environmental spatiotemporal correlation feature data;

[0107] Step S35: constructing an exogenous long-short-term memory subnet, specifically, taking the environment spatiotemporal correlation feature data as internal data, and introducing the optimized environment data and optimized mechanical data in the optimized cockpit control data set as external data, constructing an exogenous long-short-term memory subnet, and performing temporal feature processing to obtain a comprehensive output of the operating environment;

[0108] The calculation formula for constructing the exogenous long short-term memory subnetwork is:

[0109] ;

[0110] In the formula, f t is the output of the forget gate, sig(·) is the S-type activation function, W f is the forget gate weight, h t-1 is the hidden state at the previous time t-1, t is the time index, H t is the environmental spatiotemporal correlation feature data, zt is external data, b f is the forget gate bias term, I t is the input gate output, W I is the input gate weight, b I is the input gate bias term, is the candidate cell state, tanh(·) is the hyperbolic tangent function, W C is the cell state weight, b C is the cell state bias term, C t is the final cell state, O t is the output gate output, W O is the output gate weight, b O is the output gate bias term, h t is the hidden state at the current time t;

[0111] Step S36: training the work environment analysis model, specifically by combining the expert model hybrid graph convolution subnet and the exogenous long short-term memory subnet into a joint model, processing the spatial and temporal features of the environmental data and training the work environment analysis model to obtain the work environment analysis model Model WE ;

[0112] Step S37: Work environment analysis, specifically using the work environment analysis model Model WE , performing operating environment analysis based on the optimized cockpit control data set to obtain operating environment analysis reference data;

[0113] The working environment analysis reference data specifically includes working environment state mode prediction results, working environment change trend prediction results and environmental anomaly detection prediction results;

[0114] The working environment state model prediction result specifically includes the state classification of the current working environment and the specific impact weights of environmental factors;

[0115] The prediction result of the change trend of the working environment specifically refers to the change trend of the environmental variables within the next hour;

[0116] The environmental anomaly detection prediction result specifically refers to the prediction result of whether the current operating environment is abnormal or not.

[0117] By performing the above operations, in the existing working environment analysis methods, there are extremely complex environmental change factors involved in the working environment of agricultural machinery, and a large amount of environmental monitoring data needs to be detected. The existing analysis methods for the working environment basically use a single model for simple time series prediction, but this method sacrifices the evaluation of complex situations and cannot meet the subsequent technical problems of refined requirements for cockpit adjustment. This scheme creatively uses an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform working environment analysis. By constructing a mixture of expert models and graph convolution models, and introducing the operator's adjustment data as exogenous data at this stage, the connection between the results of working environment analysis and prediction and cockpit adjustment is enhanced. At the same time, the design of the expert model and the graph convolution model enhances the correlation between time and space features, and improves the accuracy and reliability of working environment analysis as a whole.

[0118] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the cockpit comprehensive adjustment is used to perform intelligent cockpit adjustment in combination with the agricultural machinery working environment and the agricultural machinery comprehensive data. Specifically, based on the working environment analysis reference data and the optimized cockpit control data set, an integrated optimized multi-agent reinforcement learning method is used to perform cockpit comprehensive adjustment to obtain cockpit comprehensive adjustment reference data. Specifically, the following steps are included:

[0119] Step S41: constructing a multi-agent environment, specifically using a multi-agent design to perform reinforcement learning agent definition, wherein the multi-agent design specifically includes a seat adjustment agent, a temperature adjustment agent, and a humidity adjustment agent;

[0120] Step S42: constructing a reinforcement learning parameter space, specifically constructing a reinforcement learning parameter combination, specifically including state space parameters, action space parameters and reward function parameters;

[0121] The state space parameters specifically include the work environment analysis reference data, operator physiological data and operator input data;

[0122] The action space parameters specifically include seat adjustment action, temperature adjustment action and humidity adjustment action;

[0123] The reward function parameters are specifically designed to calculate the reward value parameters, and the calculation formula is:

[0124] ;

[0125] In the formula, R r is the reward value parameter, is the comfort weight, specifically 0.65, So is the comfort change value, specifically calculated from the optimized operator physiological data, is the weight of the operating efficiency, with a specific value of 0.35, and Ra is the operating efficiency change value, which is specifically calculated from the operating environment analysis reference data. is the environmental suitability weight, with a specific value of 0.85, and Ev is the environmental suitability parameter, which is specifically calculated from the operating environment analysis reference data. is the adjustment frequency weight, specifically 0.15, Tr is the adjustment frequency parameter, specifically calculated from the optimization operator input data;

[0126] Step S43: Deep deterministic policy gradient optimization is used to optimize the continuous cockpit adjustment control process, specifically to construct a standard deep deterministic policy gradient optimization algorithm, optimize state space parameters and action space parameters, and optimize the learning effect of multi-agents through the experience replay mechanism;

[0127] Step S44: proximal strategy optimization, which is used to stabilize the decision results of the cockpit adjustment control as a whole, specifically, to construct a standard proximal strategy optimization algorithm and optimize the decision stability of multi-agent reinforcement learning by introducing a shear objective function;

[0128] Step S45: reinforcement learning integrated optimization, specifically, performing reinforcement learning integrated optimization through the deep deterministic policy gradient optimization and the proximal policy optimization, and performing multi-agent reinforcement learning comprehensive optimization by constructing an integrated optimization objective function to obtain a multi-agent comprehensive optimization output, and the calculation formula is:

[0129] ;

[0130] In the formula, is the multi-agent comprehensive optimization output, where is the parameter to be optimized, which is used to represent the specific cockpit comprehensive adjustment decision, E r [·] is the expected function, r is the iteration index, min(·) is the minimum function, The whole is used to represent the objective function part of the proximal strategy optimization, where is the strategy ratio, which is used to represent the reward value ratio between the new cockpit adjustment decision and the old cockpit adjustment decision, Advantage function, used to indicate the superiority of the current action, The whole is the objective function part of deep deterministic policy gradient optimization, where: is the balance coefficient, which is used as the weight for adjusting the integrated optimization. The specific value is 0.75. is the Q-value function, which is used to evaluate the current state space parameter sr and action space parameter a r The decision effect, s r is the state space parameter, a r is the action space parameter;

[0131] Step S46: training the cockpit comprehensive adjustment model, specifically, training the cockpit comprehensive adjustment model by constructing the multi-agent environment model, constructing the reinforcement learning parameter space, performing the deep deterministic policy gradient optimization and the reinforcement learning integrated optimization, and obtaining the cockpit comprehensive adjustment model Model SC ;

[0132] Step S47: Comprehensive cockpit adjustment, specifically using the cockpit comprehensive adjustment model Model SC , performing cockpit comprehensive adjustment according to the operating environment analysis reference data and the optimized cockpit control data set to obtain cockpit comprehensive adjustment reference data;

[0133] The cockpit comprehensive adjustment reference data specifically includes cockpit adjustment reference parameters, operator comfort evaluation parameters, work efficiency improvement parameters and environmental comfort index parameters;

[0134] The cabin adjustment reference parameters specifically include seat adjustment parameters, temperature adjustment parameters and humidity adjustment parameters;

[0135] The operator comfort evaluation parameter is specifically used to evaluate the operator's current cockpit comfort by detecting the operator's physiological data;

[0136] The operating efficiency improvement parameter specifically refers to a predicted parameter of operating efficiency improvement after comprehensive cockpit adjustment based on the cockpit adjustment reference parameter;

[0137] The environmental comfort index parameter specifically refers to a comfortable working environment prediction parameter of the current cockpit environment.

[0138] In step S5, the intelligent cockpit control is used to integrate the environmental analysis and the cockpit adjustment strategy to make a comprehensive control decision. Specifically, the intelligent cockpit control is performed through the operating environment analysis and the cockpit comprehensive adjustment, and based on the operating environment analysis reference data and the cockpit comprehensive adjustment reference data, to obtain the intelligent cockpit control decision reference data.

[0139] By executing the above operations, in view of the fact that there are control decision methods themselves in the existing cockpit comprehensive adjustment methods, recent technical solutions tend to optimize intelligent decisions through reinforcement learning methods. However, the comprehensive adjustment of the cockpit involves both the mechanical control of the cockpit itself and the adjustment of the physical environment inside the cockpit. Therefore, the traditional reinforcement learning method has technical problems that need to be improved in terms of accuracy and learning standardization. This solution creatively adopts an integrated optimized multi-agent reinforcement learning method to perform comprehensive cockpit adjustment. By integrating the optimization ideas of two optimization algorithms, proximal strategy optimization and deep deterministic policy gradient optimization, deep deterministic policy gradient optimization is used to improve the accuracy of continuous adjustment action decisions, and proximal strategy optimization is used to ensure that the adjustment strategy will not change drastically, thereby improving the stability of the entire learning process. At the same time, the design of multi-agents also optimizes the adaptability of reinforcement learning applied to comprehensive cockpit adjustment. By adjusting multiple levels of mechanical and physical environments, the intelligence of the cockpit comprehensive adjustment method is improved as a whole, and it also provides strong support and assistance for intelligent agricultural machinery control.

[0140] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0141] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0142] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent cockpit control method for agricultural machinery control, characterized in that: The method comprises the following steps: Step S1: Data collection, through sensor arrangement and feedback collection from agricultural machinery operators, the original data set of intelligent cockpit control is collected from the agricultural machinery operation process data; Step S2: data preprocessing, optimizing the original data set of the intelligent cockpit control to obtain an optimized cockpit control data set; Step S3: working environment analysis, used to perform basic analysis on the working environment of agricultural machinery, specifically based on the optimized cockpit control data set, using an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform working environment analysis, and obtain working environment analysis reference data, specifically including the following steps: step S31: graph structure construction; step S32: graph convolution subnet construction; step S33: expert model construction; step S34: constructing an expert model hybrid graph convolution subnet; step S35: constructing an exogenous long short-term memory subnet; step S36: working environment analysis model training; step S37: working environment analysis; Step S4: comprehensive cockpit adjustment, which is used to perform intelligent cockpit adjustment in combination with the agricultural machinery working environment and the agricultural machinery comprehensive data. Specifically, based on the working environment analysis reference data and the optimized cockpit control data set, an integrated optimized multi-agent reinforcement learning method is used to perform cockpit comprehensive adjustment to obtain cockpit comprehensive adjustment reference data, which specifically includes the following steps: step S41: constructing a multi-agent environment; step S42: constructing a reinforcement learning parameter space; step S43: deep deterministic policy gradient optimization; step S44: proximal policy optimization; step S45: reinforcement learning integrated optimization; step S46: cockpit comprehensive adjustment model training; step S47: cockpit comprehensive adjustment; Step S5: Intelligent cockpit control, obtaining intelligent cockpit control decision reference data.

2. The intelligent cockpit control method for agricultural machinery control according to claim 1, characterized in that: In step S1, the smart cockpit controls the original data set, specifically including environmental data, mechanical data, operator physiological data and operator input data; In step S2, the optimization of the cockpit control data set specifically includes optimizing environmental data, optimizing mechanical data, optimizing operator physiological data, and optimizing operator input data.

3. The intelligent cockpit control method for agricultural machinery control according to claim 2, characterized in that: In step S3, the exogenous long short-term memory neural network combined with the expert model hybrid graph convolution specifically includes an expert model hybrid graph convolution subnet and an exogenous long short-term memory subnet; The expert model hybrid graph convolution subnet is used to extract the operation environment factor graph information from the optimized environment data and the optimized mechanical data in the optimized cockpit control data set, and capture the correlation characteristics between the environmental factors through the graph convolution operation; The exogenous long short-term memory subnet is used to combine the optimized operator physiological data and the optimized operator input data in the optimized cockpit control data set as external data, and predict the future dynamic change trend of the environment by combining the correlation characteristics between environmental factors.

4. The intelligent cockpit control method for agricultural machinery control according to claim 3, characterized in that: In step S3, the step of using an exogenous long short-term memory neural network combined with an expert model hybrid graph convolution to perform work environment analysis and obtain work environment analysis reference data includes: Step S31: constructing a graph structure, specifically, constructing a working environment graph feature matrix and an environmental factor adjacency matrix according to the optimized environment data and optimized mechanical data in the optimized cockpit control data set, to obtain a working environment graph data set; Step S32: constructing a graph convolutional subnetwork, specifically, constructing three parallel standard graph convolutional neural networks based on the work environment graph dataset, and obtaining work environment related feature data through node aggregation and node relationship feature extraction; Step S33: expert model construction, specifically, constructing an expert model to adjust the weight calculation data of the working environment related feature data, and by constructing three expert models, respectively optimizing the feature extraction effects of three parallel standard graph convolutional neural networks, and obtaining the expert model optimized environment related feature data; Step S34: constructing an expert model hybrid graph convolutional subnet, specifically, constructing an expert model hybrid graph convolutional subnet through the graph structure construction, the graph convolutional subnet construction and the expert model construction, and performing a graph convolution operation by applying the expert model hybrid graph convolutional subnet to obtain environmental spatiotemporal correlation feature data; Step S35: constructing an exogenous long-short-term memory subnet, specifically, taking the environment spatiotemporal correlation feature data as internal data, and introducing the optimized environment data and optimized mechanical data in the optimized cockpit control data set as external data, constructing an exogenous long-short-term memory subnet, and performing temporal feature processing to obtain a comprehensive output of the operating environment; Step S36: training the work environment analysis model, specifically by combining the expert model hybrid graph convolution subnet and the exogenous long short-term memory subnet into a joint model, processing the spatial and temporal features of the environmental data and training the work environment analysis model to obtain the work environment analysis model Model WE ; Step S37: Work environment analysis, specifically using the work environment analysis model Model WE , based on the optimized cockpit control data set, an operating environment analysis is performed to obtain operating environment analysis reference data.

5. The intelligent cockpit control method for agricultural machinery control according to claim 4, characterized in that: In step S3, the working environment analysis reference data specifically includes working environment state mode prediction results, working environment change trend prediction results and environmental anomaly detection prediction results; The working environment state model prediction result specifically includes the state classification of the current working environment and the specific impact weights of environmental factors; The prediction result of the change trend of the working environment specifically refers to the change trend of the environmental variables within the next hour; The environmental anomaly detection prediction result specifically refers to the prediction result of whether the current operating environment is abnormal or not.

6. The intelligent cockpit control method for agricultural machinery control according to claim 5, characterized in that: In step S4, the cockpit comprehensive adjustment is used to perform intelligent cockpit adjustment in combination with the agricultural machinery working environment and the agricultural machinery comprehensive data. Specifically, based on the working environment analysis reference data and the optimized cockpit control data set, an integrated optimized multi-agent reinforcement learning method is used to perform cockpit comprehensive adjustment to obtain cockpit comprehensive adjustment reference data, which specifically includes the following steps: Step S41: constructing a multi-agent environment, specifically using a multi-agent design to perform reinforcement learning agent definition, wherein the multi-agent design specifically includes a seat adjustment agent, a temperature adjustment agent, and a humidity adjustment agent; Step S42: constructing a reinforcement learning parameter space, specifically constructing a reinforcement learning parameter combination, specifically including state space parameters, action space parameters and reward function parameters; The state space parameters specifically include the work environment analysis reference data, operator physiological data and operator input data; The action space parameters specifically include seat adjustment action, temperature adjustment action and humidity adjustment action; The reward function parameters are specifically designed to calculate the reward value parameters, and the calculation formula is: ； In the formula, R r is the reward value parameter, is the comfort weight, So is the comfort change value, which is specifically calculated from the optimized operator physiological data, is the operating efficiency weight, Ra is the operating efficiency change value, which is specifically calculated from the operating environment analysis reference data, is the environmental suitability weight, Ev is the environmental suitability parameter, which is specifically calculated from the operating environment analysis reference data, is the adjustment frequency weight, Tr is the adjustment frequency parameter, which is specifically calculated from the optimization operator input data; Step S43: Deep deterministic policy gradient optimization is used to optimize the continuous cockpit adjustment control process, specifically to construct a standard deep deterministic policy gradient optimization algorithm, optimize state space parameters and action space parameters, and optimize the learning effect of multi-agents through the experience replay mechanism; Step S44: proximal strategy optimization, which is used to stabilize the decision results of the cockpit adjustment control as a whole, specifically, to construct a standard proximal strategy optimization algorithm and optimize the decision stability of multi-agent reinforcement learning by introducing a shear objective function; Step S45: reinforcement learning integrated optimization, specifically, performing reinforcement learning integrated optimization through the deep deterministic policy gradient optimization and the proximal policy optimization, and performing multi-agent reinforcement learning comprehensive optimization by constructing an integrated optimization objective function to obtain a multi-agent comprehensive optimization output, and the calculation formula is: ； In the formula, is the multi-agent comprehensive optimization output, where is the parameter to be optimized, which is used to represent the specific cockpit comprehensive adjustment decision, E r [·] is the expected function, r is the iteration index, min(·) is the minimum function, The whole is used to represent the objective function part of the proximal strategy optimization, where is the strategy ratio, which is used to represent the reward value ratio between the new cockpit adjustment decision and the old cockpit adjustment decision, Advantage function, used to indicate the superiority of the current action, The whole is the objective function part of deep deterministic policy gradient optimization, where: is the balance coefficient, used as the weight for adjusting the integrated optimization, is the Q-value function, which is used to evaluate the current state space parameter s r and action space parameter a r The decision effect, s r is the state space parameter, a r is the action space parameter; Step S46: training the cockpit comprehensive adjustment model, specifically, training the cockpit comprehensive adjustment model by constructing the multi-agent environment model, constructing the reinforcement learning parameter space, performing the deep deterministic policy gradient optimization and the reinforcement learning integrated optimization, and obtaining the cockpit comprehensive adjustment model Model SC ; Step S47: Comprehensive cockpit adjustment, specifically using the cockpit comprehensive adjustment model Model SC , based on the operating environment analysis reference data and the optimized cockpit control data set, the cockpit comprehensive adjustment is performed to obtain the cockpit comprehensive adjustment reference data.

7. The intelligent cockpit control method for agricultural machinery control according to claim 6, characterized in that: In step S4, the cockpit comprehensive adjustment reference data specifically includes cockpit adjustment reference parameters, operator comfort evaluation parameters, work efficiency improvement parameters and environmental comfort index parameters; The cabin adjustment reference parameters specifically include seat adjustment parameters, temperature adjustment parameters and humidity adjustment parameters; The operator comfort evaluation parameter is specifically used to evaluate the operator's current cockpit comfort by detecting the operator's physiological data; The operating efficiency improvement parameter specifically refers to a predicted parameter of operating efficiency improvement after comprehensive cockpit adjustment based on the cockpit adjustment reference parameter; The environmental comfort index parameter specifically refers to a comfortable working environment prediction parameter of the current cockpit environment.

8. The intelligent cockpit control method for agricultural machinery control according to claim 7, characterized in that: In step S5, the intelligent cockpit control is used to integrate the environmental analysis and the cockpit adjustment strategy to make a comprehensive control decision. Specifically, the intelligent cockpit control is performed through the operating environment analysis and the cockpit comprehensive adjustment, and based on the operating environment analysis reference data and the cockpit comprehensive adjustment reference data, to obtain the intelligent cockpit control decision reference data.

Images